TWI641921B - Monitoring system and method for testing patterned structure measurement - Google Patents

Abstract

The present invention presents a monitoring method and system for parameter measurement of a patterned structure based on a predetermined adaptation model. The method includes: (a) providing at least one patterned structure indicating measurement data; and (b) using at least one selected verification mode on the data indicating measurement, the at least one selected verification mode includes: The predetermined factor analyzes the data and classifies the corresponding measurement results as acceptable or unacceptable, so it can be determined whether to ignore the measurement that provides more than one of the unacceptable results, or whether to modify the predetermined fit model more than one Parameters.

Description

Monitoring system and method for testing patterned structure measurement

The invention relates to a monitoring method and system for parameter measurement of a patterned structure.

Optical critical dimension (also known as "optical CD" or "OCD") measurement technology (also known as scatterometry) is a well-known and effective technology for measuring the parameters of patterned (periodic) structures. The measurement of these parameters provides a feasible metrological solution for the process control of mass-produced semiconductor devices.

OCD measurements are usually performed using an adapted procedure. According to this procedure, the theoretical model that describes the structure by measurement is used to generate theoretical data or reference data, and the latter will continue to change the model parameters until the "best fit" is found, and iteratively compares with the measured data. The parameters of the "best fit" model are considered to be equivalent to the measured parameters. The measured data (usually optical data) can be analyzed to derive information related to the geometric parameters of the pattern, including thickness, critical dimension (CD), line spacing, line width, wall depth, wall profile, etc. Contains the optical constant of the material.

The optical metrology tools used for such measurements are typically elliptically polarized and / or reflective metrology based tools. Tools based on reflection metrology usually measure changes in radiation intensity, whether returned or transmitted through the sample are unpolarized or polarized, while tools based on ellipsometry usually measure the polarity of radiation after the radiation interacts with the sample. Variety. In addition to or as an alternative to these techniques, angular analysis of light returned (reflected and / or scattered) from a patterned (periodic) structure can be used to measure parameters that define / describe structural features.

Although optical CD has demonstrated the advantages of its process control, compared to image-based metrology technology (considered to be more direct and therefore more reliable), optical CD suffers from the disadvantage of perceived reliability. One reason is that the measured spectrum cannot be easily interpreted by the human eye (as opposed to microscopic images). As a result, possible errors in the data (such as caused by a failure of the measurement tool) may go unnoticed and compromise the model fit. Another reason is that the theoretical modeling of scatterometry usually contains pre- assumptions, such as that several semiconductor materials are stable and known, that certain geometric parameters can be assumed to be constant, and so on. The actual situation that deviates from these assumptions will cause some errors in the adaptation procedure and thus affect the measurement results. In most cases, such errors are not easy to identify or quantify. The potential hazard of such errors is the sub-optimal control of the process, which may cause semiconductor yield issues.

This technical field therefore requires an innovative technique for measuring the parameters of a patterned structure that validates the measurements to determine whether a particular measurement is to be ignored or if more than one parameter of the fit model is to be modified. In this regard, it should be understood that the present invention provides verification of the measurement by using more than one verification mode on the data indicating the measurement, where such data may include raw measured data (such as before fitting) or measurement sound data ; And / or measured data conforming to the desired degree of fit of the model; and / or measurement / measurement results (such as structural parameters calculated by the fitting procedure). In the following description, the data indicating measurement of all these categories are sometimes referred to as "measured data", but the term should still be interpreted correctly as defined above.

The invention provides several key performance measurement methods, which together constitute a verification technology, which can be used as a protective measure against different types of errors in scatterometry and can mark possible measurement results that are erroneous. The present invention is based on the knowledge described below. As mentioned above, scatterometry is usually based on the combination of theoretical (analog) model data or signals with spectral (diffraction) characteristics measured by a patterned (periodic) structure. Generally, during setup (offline), the theoretical model is tested and matched to all available measurement samples, such as for calculation accuracy and / or geometric description and / or materials, etc. During the actual (connected) measurement of a new unknown sample with a set model, there are still several possible factors that may cause measurement errors. These factors may be related to the mismatch between the theoretical model and the measurement sample, such as the lack of geometric description of the stacked structure (such as missing film layers, lack of chamfers, etc.), incorrect material optical characteristics models (such as materials not considered) Variability). The assumption of fixed numerical values for several stacked parameters is not applicable to the sample (different constants or variable values). Moreover, these factors may be inaccurately related to the measured data, including systematic measurement errors (such as problems caused by at least one of calibration, pattern recognition, focusing, and / or other measurement tool subsystems), and exceeding The level of random noise that this particular tool accepts.

The present invention provides detection of erroneous measurements by scatterometry (on a measurement tool). This can be achieved by introducing a verification model that includes at least one error indicator (EI) device or a combination of multiple error indicator devices, each of which is used to calibrate more than one possible different type of measurement problem. The output data of multiple EIs can be further combined into a single evaluation verification number (VFM, Verification Figure of Merit), which is called "score", for example. By placing control limits (thresholds) on the VFM, the system can calibrate any particular measurement that does not meet the required measurement quality, and can use scatterometric metrology tools in a safer and more reliable way. The cause of the calibration measurement can be further analyzed according to the EI information, so that the user can decide whether a corrective action is required or whether the measurement result is available for process control.

The error index can be based on any single measurement or wafer statistics. For all error indicators, the confidence and score limits can be set during the recipe setting (offline) step. These limits are part of the measurement recipe and are used to calculate the evaluation verification number when the recipe is used for production measurement.

Therefore, according to a broad aspect of the present invention, the present invention provides a monitoring method for parameter measurement of a patterned structure. The parameter measurement of the patterned structure is based on a predetermined adaptation model. The monitoring method includes: (a) providing instructions at least Measurement data of a patterned structure; (b) using at least one selected verification mode on the data indicating the measurement, the at least one selected verification mode includes: analyzing the data according to at least one predetermined factor and corresponding measurement results It is classified as acceptable or unacceptable, so it can be determined whether to ignore more than one of the measurements providing the unacceptable result, or whether to modify more than one parameter of the predetermined fit model.

As mentioned above, the data indicating measurement includes at least one of the following data types: raw measured data, measured sound data, data that conforms to the degree of fit desired by the fit model, and one of the structural parameters calculated by the fit procedure The measurement results presented in the form of.

The desired degree of fitness is usually defined by the evaluation function or the fitness of the fitness factor. In some embodiments, using the selected verification mode may include analyzing multiple values of the evaluation function determined for multiple measurement locations, and determining that the multiple values of the evaluation function include at least one value that differs from other values by more than one. When a certain threshold is specified, the corresponding measurement is classified as an unacceptable result. The plurality of measurement locations may include at least one control location, which has a configuration corresponding to at least one other measurement location and is characterized by having a smaller amount of structural floating parameters. In some embodiments, using the selected verification mode includes analyzing at least two evaluation functions defined for a control location and at least one measurement location, and determining that the difference between the evaluation function of the control location and the at least one measurement location exceeds a specific For threshold values, the corresponding measurement is classified as an unacceptable result. In some embodiments, for each of the control place and the at least one measurement place, the evaluation function is used to determine a measurement result presented in the form of at least one parameter of the patterned structure.

In some embodiments, the raw measured data includes at least two data segments corresponding to at least two different measurement situations, respectively. At least one measured parameter based on the model (in accordance with a predetermined degree of fit of the raw measured data segment) can be used for each of the at least two data segments, and at least one parameter of the patterned structure is determined. In this case, using the selected verification mode may include analyzing at least one parameter of the patterned structure corresponding to at least two values in at least two different measurement situations, and determining that the difference between the at least two values exceeds a predetermined threshold , The corresponding measurements are classified as unacceptable results. The data indicating the measurement may include spectral data, wherein at least two data segments may correspond to at least two different wavelengths, respectively. In some examples, the at least two data segments correspond to an angle at which the at least two different radiations used in the measurement are incident on the structure, and / or an angle at which the radiation propagates from the structure; and the at least two data segments correspond to At least two different radiation polarities used in the measurement.

In several embodiments, the raw measured data is presented as a multi-point function of the structure's measured response to incident radiation. Using the selected verification mode may include comparing a multi-point function of the measured response with a theoretical model function that meets a predetermined degree of fit of the measured response, enabling determination of whether the multi-point function includes at least one function for the at least one measurement point Value that differs from the function value at other measurement points by more than a certain threshold.

In some embodiments, the use of the selected verification mode includes determining the number of iterative steps to achieve a desired moderate condition, and when the number of identifications is identified to exceed a predetermined threshold, the corresponding measurement is classified as an unacceptable result.

In some embodiments, the raw measured data may correspond to a measurement performed at the same measurement location and include the measurement signals continuously obtained by the measurement location, and may additionally include a serial measurement continuously measured by the same measurement location. The measured signal is a set of measured signals. Using the selected verification mode may include comparing the measured signals to each other to determine whether there is at least one measured signal that differs from other measured signals by more than a certain threshold; and / or comparing the measured signals to the set of measured signals To determine whether there is a difference between a measured signal and the set of measured signals that exceeds a specific threshold.

According to another broad aspect of the present invention, the present invention provides a monitoring system for controlling parameter measurement of a patterned structure, the monitoring system comprising: (a) a data input device for receiving a measurement indicating at least one patterned structure (B) a memory device to store at least one adaptation model; and (c) a processor device including: a adaptation device configured to use the at least one adaptation model to determine Presenting a measurement result in the form of at least one parameter of the patterned structure; a verification module including more than one error indicator device, each configured to use at least one verification mode on the measured data, the At least one verification mode includes: analyzing the data according to at least one predetermined factor and classifying the corresponding measurement as acceptable or unacceptable, and generating its indication output data, so that it can be determined whether to ignore more than one of the unacceptable measurements, Or whether to modify more than one parameter of the fitted model.

Referring to FIG. 1, a monitoring system 100 configured to control parameter measurements of a patterned structure is schematically depicted by a block diagram. The system 100 operates as a data verification system, which can be connected to the measurement unit 102 (via a line or wireless signal transmission). The monitoring system 100 is a computer system, and in particular includes functional devices listed below: a data input / output device 104, a memory device 106, a data processor unit 108, and possibly a display 110. The processor unit 108 includes an adaptation device 109, a structure (wafer) parameter (measurement result) calculator 111, and a verification module 112 including more than one error index device (generally referred to as EIi). Each error indicator device is configured to perform more than one verification mode on the data indicative of the measurement. As mentioned above, the data to be verified includes one or more of the following: raw measured data (before fitting), sound measured data, measured data described by the degree of fit with the matching model (fitness (GOF , goodness of fit), evaluation function (MF, merit function), and measurement / measurement results (that is, structural parameters calculated through an adaptation program). The error indicator device operates to process / analyze the data indicating the measurement (offline or online) and generate output data (calibration) indicating more than one potential measurement problem. In a particular but non-limiting example, each error indicator device is configured to determine / calibrate a different type of problem. As further shown in the figure, the output data from multiple error indicators can be combined into a single evaluation verification number (VFM) device 114 that characterizes the measurement quality.

Referring to FIG. 2, a flowchart of the method of the present invention performed by the monitoring system 100 is illustrated. First, more than one adaptation model is selected and stored in the memory device 106 (as part of a recipe design, such as generated by a learning model) (202). The system receives data indicative of optical measurements, and the measurements are performed at at least one measurement of more than one patterned structure (eg, multiple batches). The data indicating the measurement may be received directly from the measurement unit (online mode) or may be received from a data storage device of the measurement unit or a non-measurement unit (offline mode) (204). The received data may be raw measured data corresponding to the measurement at at least one measurement (206). It can be a series of detected optical response signals (for example from the same measurement), or it can be a multi-point function of the structure's measured response to incident radiation, or it can be a spectral feature. Additionally or additionally, the received data may include sound measured data (208), fit data (210), and measurement results (212). The processor unit 108 operates to activate the verification module (214) with more than one error index EIi to use the at least one selected verification mode on the received data to verify the data (216) and generate its instruction output data (218) . The validation mode includes analyzing the received data using more than one validation factor (threshold-based factor) and classifying the corresponding measurement as acceptable or unacceptable (220). Optionally, the data corresponding to unacceptable measurement results can be further analyzed to determine whether to ignore more than one measurement that provides unacceptable results or to modify more than one parameter of the fit model (224).

The following are some examples of the configuration and operation of the verification module 112. It should be noted that the present invention is here an example related to scatterometry measurements, in which the measured data can be presented in the form of spectral features. However, the principle of the present invention is not limited to this specific application, and the present invention can be widely used in any model-based measurement technology, that is, a technology whose measured data is interpreted by a model (adaptive program) and the sample parameters of interest Derived from model values that match the measured data for best fit. In addition, the present invention is here an example related to measurements taken on a semiconductor wafer. However, it is understood that the present invention is not limited to this particular application, and the sample / structure under measurement can be any patterned structure. The parameters to be measured may include features of the pattern (such as critical dimensions) and film thickness.

Referring to FIG. 3, which illustrates the operation of the monitoring system 100, the verification module 112 includes a so-called adaptive quality index device. The processor unit 108 receives data indicating the measurement, which is presented in the form of raw measured data corresponding to the measurements at a plurality of measurement locations, and the processor unit 108 activates a suitable quality indicator device. The latter uses the selected adaptation model (selected from the memory device 106) on the raw data (that is, executes the adaptation procedure), and for each measurement location determines an evaluation function that meets the predetermined fitness level of the raw data ( Constituting model-based measured parameters corresponding to at least one parameter of the patterned structure). Then, the error index device EIi operates to execute the selected verification mode. To this end, multiple values of the evaluation function for multiple measurement locations are analyzed. If there is at least one evaluation function value that differs from other values by more than a predetermined threshold value, the error index device EIi will classify the corresponding measurement result as unacceptable and generate corresponding output data; if not, the processor unit 108 will use This evaluation function calculates the structural parameters (measurement results).

As mentioned above, the fitness evaluation function can be calculated as a function of the difference between theoretical and experimental diffraction characteristics. The fitness evaluation function is usually used as the minimization parameter in the fitness program, which means that the model parameters are repeatedly modified and the fitness evaluation function is repeatedly calculated until the fitness evaluation function reaches the minimum. Once the best fit between the model and specific measurement data has been reached, the final value of the fit evaluation function represents the residual mismatch error that cannot be described under the assumptions of the current model. If the residual value of the fitting evaluation function exceeds the standard degree defined in the model setting, it indicates that there may be a problem, for example, the measured data violates one of the model assumptions.

In this regard, reference is made to FIGS. 4A to 4D. FIG. 4A shows a schematic cross-sectional view of a typical periodic structure (only three cycles are shown) for performing scatterometry, which includes a plurality of uniform film layers and a periodic photoresistance line array on the top layer. At a measurement location, the optical characteristics of one of the stacked materials have been intentionally modified, while the theoretical model uses fixed material characteristics.

FIG. 4B shows the measured values of the cross-sectional height of the structure of FIG. 4A with and without floating material characteristics (curves G1 and G2, respectively). Here, the measurement results are presented in the curve G2, which shows that the value of the height of the section becomes obviously lower at one of the measurement locations.

FIG. 4C depicts the results of the evaluation function at multiple points at three measurement locations. The material properties of one of the film layers in the structure were deliberately modified at a measurement point, as shown by the higher value of the evaluation function obtained when n & k (material properties) remains fixed (block S). When the n & k value is floated, the degree of the evaluation function returns to the same level as the other two (diamond D). As shown in FIG. 4C, the fitness evaluation function clearly shows that the value at a particular measurement is significantly higher than elsewhere, and therefore the suspicious measurement is correctly calibrated. By incorporating variable material properties into the model (G1, D), the correct profile height measurements and the appropriate evaluation function values can be retrieved.

FIG. 4D shows the measured (correlation) values of the material properties of the change layer (similar to the example of FIG. 4C), which is reconstructed by an adaptation procedure under the condition with floating variables.

Referring to FIG. 5, which illustrates the operation of the monitoring system 100, the verification module 112 includes a so-called single measurement noise indicator device. The data to be verified can be presented in the form of unprocessed measured data, which corresponds to the measurement performed at the same measurement location, and includes a string / a group of measured signals continuously measured from the measurement location, or is measured by the group The set of measured signals forms the measured signals. The error indicator equipment EIi uses this verification mode for raw measured data. The verification mode may include a comparison of the measured signals with each other to determine whether there is at least one measured signal that differs from other measured signals by more than a specific threshold. In addition or in addition, the verification mode includes a comparison between each measured signal and the collective measured signal to determine whether the measured signals include at least one measured signal and the collective measured signal differs by more than a specific threshold.

In this regard, it should be understood as follows. The repeatability of measurement results (after fitting) is probably the most commonly used quality measure in offline data analysis and tool certification. Low repeatability is an indicator that the tool may have problems (such as high noise or stability issues), but it may also indicate a change in one of the model assumptions. For example, if it is assumed that a fixed parameter is changed, the solution will not only become less accurate but also unstable, thus reducing the repeatability of a specific parameter. Because iterative data collection steps take a lot of time and have no value for production, repeatability testing procedures are usually performed only in specialized tests. However, if the measured data noise can be measured, there are several options to assess repeatability with minimal output shock. To evaluate noise in real time, one of the following can be used on the actual target (location, wafer).

In several cases, the data to be interpreted can be (or has been) collected in a series of short exposures rather than a single long exposure. Different exposures are usually averaged to minimize noise. However, if read separately, short exposures can be used to evaluate the noise remaining in the average result. Taking the assumption of Gaussian random noise as an example, the average noise RMS of the average is the noise RMS of each single measurement in the average divided by the square root of the (known) number of single measurements.

Another example is based on the need for a statistical indicator (such as for a wafer or a batch) rather than a specific measurement in many situations, and thus repeatability can also be achieved by using bootstrapping methods. To evaluate. More specifically, the second measurement can be collected at several locations (such as one on each wafer), and the differences between the two measurements are recorded over time and accumulated to obtain the representative value of the typical noise. . For example, if a fully loaded batch (25 wafers) is measured and a single location (die) of each wafer is measured twice, then the output impact on the overall system is minimal. Establish reliable measurements of spectral noise for each batch.

During formula setting, the impact of spectral noise on formula performance is defined by sensitivity and correlation analysis and stored in the formula for subsequent use during measurement to calculate the impact of measured spectral noise on all floating parameters.

Referring to FIG. 6, which illustrates the operation of the monitoring system 100, the verification module 112 includes a control error indicator also referred to as a “parallel interpretation indicator device”. In this example, the received data includes raw measured data indicating optical measurements performed at multiple (generally at least two) measurements including at least one test / control. For semiconductor wafers, the test location is usually located at the edge of the wafer, while the "actual" measurement location is located at the patterned area (die area) of the wafer.

In several embodiments, the control point indicator device is configured to perform a verification model on measured data fragments of the test point and more than one actual measurement point. When it is understood that if more than one actual measurement location is taken into account, the actual measurement locations may be on the same wafer or on similar locations of multiple wafers in different batches. In several other embodiments, the control point indicator device is configured to execute a verification model on measured data fragments located at different tests in the same wafer or similar wafers of different batches.

Therefore, the indicator unit receives the raw measured data of the second / second segment, including the measured data from the control / test location and from the "actual" measurement location in the patterned area of the wafer, or Measured data. The processor unit by using the selected adaptation model for the measured data segment (that is, performing an adaptation procedure for each raw measured data segment) and determining the value of the individual (in line with the best fit) evaluation function 108 operates to process the measured data fragments.

Next, in several embodiments, the verification model may include comparing evaluation function values between different locations and determining whether the difference therebetween exceeds a predetermined threshold corresponding to an unacceptable condition for the measured data at the actual measurement location. In several other embodiments, the evaluation function value can be used to determine the individual values (measurement results) of the specific parameters of the structure, and then the verification model is used for the measurement results (not the evaluation function values) for different locations to determine the Whether the difference between them exceeds a predetermined threshold, it is unacceptable measured data for the actual measurement.

In this regard, pay attention to the following. One reason for scatterometry errors may be related to the ability of the fit model to compensate for unmodeled differences caused by multiple interrelated floating parameters. For this reason, there may be measurement deviations that are not noticed because they are well-fitted. To overcome such difficulties, additional measurements can be used at the control to verify the quality of the measurement. Such a control location may be a relatively simple location that does not require many floating parameters, such as unpatterned (pure) locations, where all the film layers in the measurement point area at the test location are uniform. The smaller amount of floating parameters makes such positions more sensitive to changes in fixed values, such as material thickness, its optical characteristics, and / or uniformity.

Therefore, the error indicator at the control point may include a fitting evaluation function analyzer similar to that shown in FIG. 3 and / or a structural parameter analyzer (measurement result analyzer) as illustrated in FIG. 6. In the latter case, the error index can use threshold technology for common parameter values, that is, determine whether the difference between the common parameter values at the control and actual measurement locations exceeds a predetermined threshold.

For this reason, referring to FIG. 7, the control points (curve H2) of the material characteristics floating are plotted against the measurement positions (curve H1) of the fixed material characteristics by measuring in three different batches of Lot1, Lot2, and Lot3. Thickness measurement (structural parameter). For fixed n & k values, the differences between these batches are more significant. In addition or in addition, the error index can be used to identify the rapid changes in the values of different measurement parameters in the control. In order to minimize the impact of throughput (TPT, throughput), before the assumption that the error index of the control area is calibrated to identify non-local problems, the sampling plan of the control area can be quite sparse. .

Referring to FIG. 8, which illustrates the operation of the monitoring system 100, the verification module 112 includes a so-called differential measurement (data splitting) device. As shown in the figure, the processor unit 108 includes a data splitting device DS, which splits (before adaptation) the raw measured data into two raw data fragments / parts A and B ( (Generally at least two data fragments). For example, it may be different wavelengths of incident light, and / or different polarities of incident light, and / or different angles of incidence. More specifically, the raw measured data can be presented in the form of spectral characteristics measured for a group of discontinuous wavelengths in a specific wavelength range, and the data splitter respectively targets odd numbers from the group of discontinuous wavelengths. The even-numbered wavelengths provide two spectral feature parts of the spectral feature. The model adaptation is applied to each raw data part, and the same structural parameters (measurement results) are calculated for the measured data parts A and B. Then, the error indicator operates to compare the parameter values and determine whether the difference exceeds a certain threshold.

The principle of this embodiment is related to the following. A frequently asked question is not only about the existence of possible deviations, but about the possible magnitude of the deviation (error) for the parameter of interest caused by the problem at hand. One possible way to obtain this effect assessment is to examine the consistency of interpretation by using what can be called "division and comparison". Generally, in all types of scatterometry tools, multiple data points are used together to establish a small number of floating parameter values, and the number of data points is much larger than the number of parameters (data points can be at different wavelengths, angles of incidence , Polarity, or any combination). It is common practice to use all possible data points to produce a measurement with the best repeatability and accuracy. To verify accuracy and assess interpretation errors, the results can be compared to several benchmarks. Such benchmarks can be obtained from the measured data itself, by dividing the data into two parts and comparing the results. The two sets of data should preferably have different characteristics but similar repeatability. As mentioned above, possible splitting implementation schemes are: for example, choosing different polarities, splitting the wavelength range somewhere, using data from different angles, or some combination of the above. After splitting the complete data set as described above, three interpretations can be performed-once for the full data set and once for each group. When a complete group interpretation can be provided, the difference between the two groups can be used as an error indicator for the measurement problem.

When note that in most cases, even under normal conditions, there may still be some differences between the results of the two sets of data. However, because different parts respond differently to changes in measurement errors or model assumptions, the extent of these differences may increase, so thresholds for calibration of abnormal performance can be set. It should also be noted that the verification mode can actually be implemented by multiple such error indicators corresponding to multiple floating parameters, so that more representative parameters can be selected or all parameters can be tracked, each with its own threshold Degree (it can be studied according to the preliminary data set for the set period). In addition, it should be noted that the error indication obtained in such verification is presented in units of measured parameters (usually nanometers or degrees). Although the similarity of this unit does not guarantee that the error index provides the true information of the error-bar of the measurement, when the two sets of measured data are properly selected, the error index may be slightly less than the true accuracy error. Proportion, at least in magnitude. In addition, if the interpretation time is not a limiting factor, the same data can be divided into arrays to obtain additional error indicators, thereby maximizing the sensitivity of the entire error indicator module.

Figure 9 depicts the principle of the data splitting method in a curve. In this figure, the measurement of the pattern side wall angle (SWA) is presented in the form of two groups of measured data—curves P1 and P2. In this non-limiting example, it is for three batches Lot1 , Lot2, and Lot3 correspond to different polarities of incident light. As shown in the figure, the two data snippets differ significantly only almost on Lot2 measurements. Therefore, under this condition, two sets of data (different polarities in this case) can be clearly seen in both Lot1 and Lot3. However, in Lot2, where the material properties are incorrectly fixed, these two sets of measurements are measured separately. The difference between the values of the sidewall angles becomes significant, so thresholds can be used to calibrate the problem.

10, the operation of the monitoring system 100 is illustrated, in which the verification module 112 includes a so-called residual misfit analyzer RMA. According to this embodiment, the problem situation is characterized by analyzing the spectral shape of the residual (such as the residual error versus the wavelength). In any practical situation, due to all types of measurement accuracy issues or modeling approximations or inaccuracies, different parts of the signal (such as different spectral bandwidths) are fitted to different degrees. Therefore, the residuals usually have a clear, distinct spectral shape that exceeds the level of random noise. The spectral shape of the residual can thus be used as a marker for measurement, and anomalies can be identified by quantifying different parameters in the case of the markers that have produced some abnormal behavior. For example, a simple method is to divide the data into several parts, such as different spectral bandwidths, different polarities, etc. as described above, and calculate the mean square difference between the best fit and the measurement separately for each part . By tracking the error of a specific part (such as the error of the UV part in the spectrum), or its function (such as the ratio of UV error to IR error), you can get a meaningful error index that can be studied during the set period and compared with the threshold Compared with the calibration to show abnormal performance. It can also use more complicated techniques, such as using different conversion, motion, or classification techniques to identify changes in the residual signal compared to typical markers learned during the set period.

Referring now to FIG. 11, which illustrates another embodiment of the present invention, the verification module 112 includes an adaptive convergence measurement device FCMU. This indicator device determines the possible mismatch between the measured data and the model under the dynamics of the convergence algorithm. The inventors know that when there is an error in the measured data or model, it takes longer than usual to converge from the starting point to the best fit, which means that the required number of iterations is greater. Therefore, whenever the convergence value is greater than the number of statistical verifications of the iterations defined during the recipe setting, a possible problem can be calibrated. More specifically, the processor unit 108 operates to apply the selected adaptation model to raw measured data (measured radiation response to incident radiation at more than one measurement location), where the adaptation procedure includes more than one iteration Steps until you reach the best fit. This verification mode includes analyzing the number of iterative steps applied to achieve the best fit condition to determine whether this number exceeds a certain threshold, and thus identifying a condition that is unacceptable to the measured data.

Reference is now made to Fig. 12, which presents yet another embodiment of the present invention, wherein the verification mode uses an analysis that measures sound data. The measurement of health indicators should preferably cover both hardware (HW) performance and possible mismatches between samples and measurement formulations. Any measure of soundness can indicate a possible impact on measurement reliability. The input contains parameters that characterize the measurement system during the measurement, such as illumination intensity and stability, quality of pattern recognition, positional deviation of the measurement from the target (e.g. due to possible mismatch between the sample and the measurement recipe), measurement Focus quality and more. In this situation, when the quality score drops below a predetermined value, it can be used as an indicator of error.

It should also be noted that the error indicator module (tool soundness indicator) THI can be configured to determine the credibility and score limits of the measured parameters. During the formulation process, the sample groups representing the differences in the actual process were studied. According to the results, statistical confidence limits are set for all measured parameters for subsequent use in calculating the measurement scores of the production samples. Score limits are also set to indicate the possible range of parameters or formula boundaries under which the formula is valid.

Note that all or several of the above-mentioned exemplary error indicator devices can be used as a single position indicator as well as wafer statistics indicators (wafer average, range, standard deviation, etc.). There are additional indicators that can only be used at the wafer level.

Referring to FIG. 13, it illustrates a verification module configured to determine the reliability and score limit CSL of a measured parameter by monitoring a cross-wafer mark. The measured parameters usually have typical process-related cross-wafer space markings, such as constants, center-to-edge, and so on. Changes in space markers can be an indicator of a problem. For example, fixed parameters with spatial markers change and produce non-constant spatial markers in generally consistent parameters, as shown by cross-wafer differences. Moreover, the interference of ring-shaped symmetry space marks that are common in many process steps can be identified by breaking the cyclic symmetry. By knowing the cross-wafer differences from the setup stage, it is possible to determine whether there is a single outlier (trigger) measurement or if all measurements on the wafer are unreliable when one of the measurement points is far from the wafer intermediate value or deviates from the expected distribution . Therefore, the measured data can be provided in the form of a multi-point function of the structure's measured response to incident radiation, and the verification mode includes a comparison of the multi-point function of the measured response and a predetermined fit to the measured response based on a theoretical model Function to determine whether the multi-point function contains at least one function value for at least one measurement point and the function values of other measurement points differ by more than a specific threshold.

As shown above, to provide a consistent monitoring system, all or at least part of the above-mentioned exemplary error indicator devices can be combined into a single evaluation verification number (VFM) available to the user. VFM can be calculated for each measurement sample (position or die) and for wafers. A grain (measurement or grain) VFM can incorporate all or several error indicators for that particular measurement or grain. Wafer VFM can combine the VFM of all dies and add wafer-related statistics and indicators of wafer mark errors.

Fuzzy logic can be used to combine all error indicators, first the degree of the position or die, and then the degree of the wafer. This can be a rule-based combination, and a threshold (or two thresholds, minimum and maximum, depending on the situation) is assigned to each error indicator. You can also choose to define a warning zone near this threshold for each indicator. Each error indicator is then evaluated and assigned one of three individual states: pass ("green light"), failure ("red light"), or warning ("yellow light"). Knowing the status of all individual error indicators, rules can be defined in the spirit of the well-known Western Electric Rules. For example, if at least three indicators are in the red zone, the total The result is "failure". If at least one indicator is "failure" and at least two indicators are "warning", the total result is "warning" and so on.

In the spirit of fuzzy logic, a value between 0 and 1 can be defined for each error indicator, where the value of the failure zone is 0, the value of the pass zone is 1, and the value between is monotonic. (Such as linear). For example, the total value can be obtained by first combining all values for different error indicators, and then comparing the result with a threshold value (used to define a warning message). The combination of the completed error indicators can be based on the weight of each indicator defined by the user, and it allows the user to define the most important or relevant indicators. Such logic can also be implemented in multiple layers. For example, the error indicators are grouped together, and the fuzzy logic values of all members of each group are summed; the threshold degree is defined for each group, and the correlation between the group total and the threshold is defined Assign fuzzy logic values; add group results and compare with threshold to get final value. The potential advantages of such systems over individual rule-based approaches can be demonstrated in complex situations and in situations where the need to evaluate seemingly conflicting information.

And, a technique based on a learning system can be used, which plots the phase space of normal error index values and learns how to distinguish between normal and abnormal performance. Examples of normal performance are obtained from qualified data sets obtained through the use of sound tools and qualified measurement processes. To provide training groups for abnormal performance, you can intentionally offset the model's fixed parameters or intentionally skew the measured values in a mode similar to known hardware problems, such as adding random gain or random noise. Once two sets of data are provided, a learning system (such as an artificial neural network) can be trained to distinguish between good and bad measurements. This system is then used to classify each new measurement during execution.

100‧‧‧ monitoring system

102‧‧‧Measurement unit

104‧‧‧Data input / output equipment

106‧‧‧Memory Device

108‧‧‧ processor unit

109‧‧‧Fit equipment

110‧‧‧ Display

111‧‧‧ structure parameter (measurement result) calculator

112‧‧‧Verification Module

114‧‧‧ Evaluation Verification (VFM) equipment

EIi‧‧‧Error indicator equipment

DS‧‧‧Data Splitting Equipment

RMA1‧‧‧ Residual Unfit Analyzer

FCMU1‧‧‧ Equipped with appropriate convergence measurement equipment

THI‧‧‧Tool health indicators

CSL1‧‧‧Credibility and score limits

In order to better understand the subject matter disclosed herein and exemplify how it is actually implemented, embodiments will now be described by way of non-limiting examples and with reference to the accompanying drawings, in which:

1 is a block diagram of a monitoring system for controlling optical measurement of a patterned structure according to the present invention;

2 is a flowchart illustrating a method for controlling optical measurement of a patterned structure according to the present invention;

Figure 3 illustrates the operation of the monitoring system of Figure 1, where the error indicator module includes a so-called adapted quality indicator device;

4A shows a schematic cross-sectional view of a typical periodic structure;

4B shows the measured values of the cross-sectional height of the structure of FIG. 4A with and without the characteristics of a floating material;

FIG. 4C depicts the evaluation function results of multiple points at three measurement locations in the structure of FIG. 4A;

FIG. 4D presents the measured values of the material characteristics of the changing layer in the structure of FIG. 4A, which is reconstructed by the adaptation procedure in the case of having a floating variable;

FIG. 5 illustrates the operation of the monitoring system of FIG. 1, where the error indicator module includes a so-called single measurement noise indicator device;

Fig. 6 illustrates the operation of the monitoring system of Fig. 1, in which the error indicator module includes an error indicator at the control point, also known as a "parallel interpretation indicator device";

Figure 7 depicts the measurement of the thickness of a film layer with changes in material properties versus fixed material properties, as measured in unpatterned areas (test areas) in three different batches;

FIG. 8 illustrates the operation of the monitoring system of FIG. 1, where the error indicator module includes a so-called differential measurement (data splitting) device;

Figure 9 depicts the principle of the data splitting method with a curve;

FIG. 10 illustrates the operation of the monitoring system of FIG. 1, where the error index includes a so-called residual misfit analyzer;

FIG. 11 illustrates an embodiment of the present invention, in which the error index module includes an adaptive convergence measurement device;

FIG. 12 illustrates an embodiment of the present invention, in which the error indicator module uses a verification mode based on tool sound data analysis; and

FIG. 13 illustrates an embodiment of the present invention, in which the error indicator module determines the credibility and score limit of the measured parameters by monitoring the cross-wafer marks.

Claims (23)

A method for measuring the parameters of a patterned structure. The measurement of the parameters of the patterned structure is based on a predetermined fit model. The method includes: (a) using one or more optical measurements on at least one patterned structure, and Generating measured data indicating the one or more optical measurements in the at least one patterned structure, the measured data indicating the one or more optical measurements including at least raw measured data; (b) based on the raw Analysis of the measured data and the data to be verified in the measured data to select at least one verification mode, and using the selected at least one verification mode on the raw measured data, and using the at least one verification mode Including: analyzing the raw measured data using data based on at least one predetermined factor to classify corresponding measurement results of the raw measured data as acceptable or unacceptable; and when one or When more are classified as unacceptable results, it is determined whether or not one or more of the optical measurements providing the unacceptable results are to be omitted from the parameter measurement of the patterned structure. And more, or whether to modify one or more predetermined parameters of the fit of the measured parameters of the model structure of the pattern of use. For example, the method for parameter measurement of a patterned structure according to item 1 of the patent application scope, wherein the raw measured data includes two or more data segments, and the raw measured data indicates that at least two measurement locations are adopted. Such optical measurements. For example, the method for parameter measurement of a patterned structure according to item 2 of the patent application, wherein the raw measured data indicating the optical measurements include at least one of the following: (i) measured data from at least one control location , The at least one control place includes a control patterned structure having a configuration corresponding to one of the patterns in the patterned structure to be measured; (ii) measured data from at least one actual measurement place, the At least one actual measurement location is located in a patterned area of the structure; or (iii) measured data from at least two different controls, each of which has a different patterned structure, each having a pattern corresponding to Configuration of one of the patterns in the patterned structure. For example, the method for parameter measurement of a patterned structure according to item 2 of the patent application scope includes processing the raw measured data by: using a selected fit model on each data segment, and determining correspondence An individual evaluation function value under an optimal fit condition. For example, the method for parameter measurement of a patterned structure according to item 4 of the patent application, wherein the step of using the selected at least one verification mode includes: comparing values of complex evaluation functions for different measured data fragments, It is determined whether the difference between them exceeds a predetermined threshold value, which is defined by a control limit of the verification mode and corresponds to an unacceptable condition of the raw measured data at the individual measurement location. For example, the method for parameter measurement of a patterned structure according to item 4 of the scope of the patent application includes using the evaluation function values to determine individual complex values of specific parameters of the structure, and using the complex values on the complex values of the specific parameter. The at least one verification mode is selected to determine whether the difference between the values of the specific parameter exceeds a predetermined threshold value, the predetermined threshold value is defined by a control limit of the verification mode, and corresponds to an actual measurement Unacceptable measured data. For example, the method for parameter measurement of a patterned structure according to item 2 of the patent application, wherein the plurality of measurement locations include at least one control location, and the at least one control location includes a control structure having the following configuration: corresponding to at least one other measurement location And is characterized by a smaller number of patterned structure floating parameter configurations compared to the at least one other measurement location. For example, the method for parameter measurement of a patterned structure according to item 7 of the application, wherein the step of using the selected at least one verification mode includes: analyzing at least two evaluation functions, and the at least two evaluation functions are respectively for A control office and at least one measurement office; and once it is determined that the difference between the evaluation functions of the control office and the at least one measurement office exceeds a specific threshold value defined by a control limit of a verification mode , The corresponding measurement is classified as an unacceptable result. For example, the method for parameter measurement of a patterned structure according to item 8 of the patent application scope further includes using an evaluation function for each of a control place and at least one measurement place, and determining at least one parameter of the patterned structure. A measurement of the form. For example, the method for parameter measurement of a patterned structure according to item 1 or scope of the patent application, wherein the raw measurement data is a complex measurement corresponding to the complex measurement of the structure, and the structure responds to the measurement of incident radiation. Rendered in the form of a multipoint function. For example, the method for parameter measurement of a patterned structure according to item 10 of the application, wherein the step of using the selected at least one verification mode includes: the multi-point function of the measured response and a theoretical model-based function For comparison, the theoretical model-based function corresponds to a predetermined degree of matching with the measured response; and determining whether the multipoint function includes at least one function value of at least one measurement point, which is a value different from the function values of other measurement points. Exceeds a specific threshold defined by a predefined control limit of the validation mode. For example, the method for parameter measurement of a patterned structure according to item 1 or 2 of the patent application scope, wherein the unprocessed measured data corresponds to a complex optical measurement performed at the same measurement location, and includes continuously obtained from the measurement location Plural measured signals. For example, the method for parameter measurement of a patterned structure according to item 12 of the application, wherein the step of using the selected at least one verification mode includes comparing the plurality of measured signals with each other to determine whether there is at least one measurement. The resulting signal has a value that differs from other measured signals by a particular threshold value defined by a predefined control limit of the verification mode. For example, the method for parameter measurement of a patterned structure according to item 12 of the application, wherein the unprocessed measured data further includes a set of measured signals formed by a series of measured signals continuously measured at the same measurement location. For example, the method for parameter measurement of a patterned structure according to item 14 of the patent application, wherein the step of using the selected at least one verification mode includes comparing the plurality of measured signals with each other to determine whether at least one measurement exists The resulting signal has a value that differs from other measured signals by a particular threshold value defined by a predefined control limit of the verification mode. For example, the method for parameter measurement of a patterned structure according to item 14 of the application, wherein the step of using the selected at least one verification mode includes comparing the plurality of measured signals with the set measured signals to It is determined whether there is at least one measured signal whose value differs from the measured signal of the set exceeds a specific threshold value defined by a predefined control limit of the verification mode. For example, the method for parameter measurement of a patterned structure according to item 16 of the application, wherein the step of using the selected at least one verification mode includes comparing the plurality of measured signals with the set measured signals to It is determined whether there is at least one measured signal whose value differs from the measured signal of the set exceeds a specific threshold value defined by a predefined control limit of the verification mode. For example, the method for parameter measurement of a patterned structure according to item 1 or 2 of the application, wherein the unprocessed measured data includes a spectral characteristic of a measured response of the structure to one of incident radiation. For example, the method for parameter measurement of a patterned structure according to item 18 of the application, wherein the spectral characteristic corresponds to the measured response of a specific set of discontinuous wavelengths in a specific wavelength range. For example, the method for parameter measurement of a patterned structure according to item 1 or 2 of the patent application scope, wherein the measured data indicating the one or more optical measurements further includes at least one of the following: measuring sound data, corresponding to the Data on the desired degree of match of the predetermined fitting model, and a measurement result in the form of one or more structural parameters calculated by a fitting program. For example, the method for parameter measurement of a patterned structure according to item 1 or 2 of the patent application scope further includes providing data indicating an expected degree of matching with the predetermined fitting model, and the matching expected degree is determined by an evaluation function. Or as defined by a moderation factor. For example, the method for parameter measurement of a patterned structure according to claim 21, wherein the step of using the selected at least one verification mode includes: analyzing the complex value of the evaluation function determined for each of the complex measurement locations; and When it is determined that the complex value of the evaluation function contains at least one value, the difference between the complex value and the others of the complex value exceeds a specific threshold value defined by a control limit of the verification mode, the corresponding measurement is classified as impossible Accept the results. A system for controlling parameter measurement of a patterned structure, the system comprising: (a) an optical measurement unit configured and operable to use one or more optical measurements on a patterned structure and generate an indication of the One or more optically measured measured data, the measured data including raw data in the form of one or more measured signals; (b) a data input device for receiving an indication of at least one of the patterned structures; The measured data of one or more optical measurements; (c) a memory device for storing at least one adaptive model; and (d) a processor device including: an adaptive device configured and capable of Operated to utilize the at least one adaptation model and use an adaptation program on the measured data; a verification module including one or more error indicator devices, which are configured and operable to be used in the At least one verification mode is used on the measured data. The at least one verification mode is selected based on the analysis of the raw data. The at least one verification mode includes: analyzing the measured data based on at least one predetermined factor to convert a The corresponding measurement is classified as acceptable or unacceptable and produces output data indicating the corresponding measurement; and when one or more of the measurements are classified as unacceptable, it is determined whether or not to remove from the patterned structure. One or more of the measurements that are unacceptable are ignored in the parameter measurement, or one or more parameters of the at least one adapted model used in the parameter measurement of the patterned structure are to be modified.